Learning From Your Own Traces

Can an agent learn from its own verified traces, while preserving everything it already knew?